Figure 5.
Static and dynamic learning within a 2-POST network. (a) Schematic illustration of a perceptron network with a 3 × 3 PRE layer and 2 POSTs, with 18 synapses between the PRE and POST layers. Inhibitory synapses connect the 2 POSTs to reduce the internal potential of one POST when the other POST fires. (b) Patterns submitted during the first phase (top and bottom bars for static learning) and sequence of 4 pattern shifts for the dynamic learning phase. (c,d) Synaptic weights for POST1 and POST2 at the end of each sequence of learning indicated in (b). (e,f) Time evolution of the synaptic weights 1/R for POST1 and POST2 during the 5 phases of the dynamic learning. Pattern weights (red) and background weights (blue) tend to LRS and HRS, respectively.